This invention relates to an electronic field for realizing a multi-media information processing system which is used for storing and transmitting compressed information indicative of two-dimensional images, figures, etc., or for realizing a highly intelligent processing system which is used for extracting and recognizing the features of images, figures, etc.
It has been requested that the integration scale and operation speed of LSIs for use in computers or signal processing systems be increased. The conventional LSIs have been mainly used to form digital circuits. When an analog circuit is necessary, an AD or DA converter is generally interposed as an interface between the analog circuit and a digital circuit. The increase of the integration scale and operation speed of digital LSIs, however, will inevitably increase the power consumption. In other words, the increase of the integration scale and operation speed is limited in light of power consumption.
The conventional image and figure processing systems employ a structure in which information is input by an image sensor, then converted to a digital signal by an AD converter, and processed by a digital signal processor or a microprocessor. Although an information processing system using a digital circuit is most excellent in accuracy and stability, it requires multiple elements since it uses each transistor as a switching element. Increase of the integration scale and the operation speed of the system inevitably increases the energy consumption and limits the performance. Since the information processing architecture in the case of using a digital circuit is basically sequential processing, which is not suitable to parallel processing of two-dimensional information or calculation of multiple inputs employed in a neural network.
On the other hand, the analog circuit uses a transistor as a linear element such as an amplifier, and processes an analog signal whose level continuously varies with the passing of time. The analog circuit performs calculation on the basis of physical basic laws, and hence can realize the same function as the digital circuit, using 1/10 or 1/100 of the elements used in the digital circuit. Further, the former is excellent in relatively roughly grasping a state and judging it as a result of comparison, and is suitable for parallel operation. However, it is hard for the analog circuit to process information with high accuracy. In now available apparatuses each equipped with both analog and digital circuits, AD or DA conversion limits their performance, and noise (cross talk) leaking from the digital circuit to the analog circuit also limits their performance.
Image or figure processing sensors, which have been regarded as more and more important, are mainly provided in the form of CCD devices which treat input information as an analog charge amount. However, these devices are produced in a special process, and hence not suitable to integration with CMOS logic circuits for signal processing.